Abstract
While speech perception amidst competing talkers is well-studied, the perception of polyphonic music remains less explored. Pitch differences aid in source segregation, yet reductions in harmonicity have relatively little effect on speech intelligibility in such conditions. We hypothesized that source identification and segregation in music would rely more on harmonicity, given the central role of pitch in music and fewer alternative segregation cues, such as temporal incoherence. Polyphonic vocal passages from a female singer were generated from four-second MIDI excerpts of existing music. Inharmonic versions were created by introducing random frequency deviations within the first 30 harmonics of each note. Timbral cues were minimized by using a single source for all voices. Experiment 1 tested participants' ability to estimate the number of voices in passages with one to five simultaneous voices. Experiment 2 tested participants' ability to follow one voice within a two- to five-voice passage. Both experiments varied the degree of inharmonicity. Accuracy decreased with increasing inharmonicity and voice count but remained above chance in all conditions. The results confirm the importance of harmonicity in source segregation within music but show that some segregation abilities remain, even when timbral cues are removed and harmonicity is severely degraded.